Validation · fourteen rules · run on every load
What this desk will not print.
Three structural breaks sit in the IF.Data extract: the credit book changes definition at 2025 Q1, margins reset at 2025 Q4 for the payments cohort, and some provision charges arrive with the wrong sign. Every ratio built on a broken denominator is wrong by construction, so the harness runs first and the numbers come second.
01 · Register
Fourteen checks, every failure, and what each one changed
A rule fires on a print, not on a bank. “Restated” means the value is suppressed or the series is split; “documented” means the value stands and the label around it is corrected. Open any bank to see the same findings beside its numbers.
02 · Quality
Which banks you can quote without a caveat
| Institution | Type | Quality | Restated | Gaps | Notes | First finding |
|---|
03 · Coverage
What exists, quarter by quarter
Gate: a field below 80% usable coverage at the reference quarter is not offered as a screen metric.Fill is the share of the 34 institutions with a usable print after the harness. A darker cell is fuller coverage. 2024 Q4 is missing for the whole credit block — the hole is kept, never interpolated.
04 · Dictionary
Every field, its numerator, and where it comes from
Fields marked absent are published by IF.Data but not joined here. They are listed with their definition so that a coverage gap reads as a missing fetch, not as a missing number in the world. Basel, CET1, RWA and RAROC were joined on 2026-10-20 from report 5, and the bank-level defaulted book from report 130; they live in separate feeds, carry their own coverage notes on the Capital lab and Credit & risk screens, and the defaulted series starts in 2025 Q1 because the source does not publish it earlier.
05 · Decisions
What changed, and what it cost
06 · Open
What the harness still cannot defend
Bank-level NPL90 and market prices are still missing. Capital and RWA are now joined from IF.Data report 5 — 425 points across 13 quarters, with 17 gaps kept as gaps — so the capital view is leverage-adjusted. Asset quality remains system-level because the per-bank NPL join has not been done, and no valuation ratio is shown without a point-in-time price history.
Banco do Brasil’s margin is not explained. NIM falls from 2.6% to 0.7% across eight quarters while its credit book grows, and no break in the source accounts for it. The harness flags it; the income lab shows the resulting residual rather than smoothing it.
Snapshot as of —, retrieved —. A refresh may only publish when this harness returns zero errors on the current quarter; otherwise the previous snapshot stands.